Inconsistent results as a function of methodological differences across studies have sometimes been cited as challenging the classification of the audi-tory sensory gating deficit as a core feature of schizophrenia psychopathology.
The present report evaluated six common approaches within the same experi-mental protocol and for both chronic and first admission patients compared to healthy controls in order to determine, in the same data set, how substantial such discrepancies are.
Per hypothesis 1, a gating deficit in M50 in schizophrenia patients was observed consistently across gating quantification methods. The gating ratios varying between .59 and .73 for CHR and FA vs. and between .48 and .54 for HC are within the range reported in meta-analyses (de Wilde et al., 2007; Patterson et al., 2008), although effect sizes between .43 and .70 are less than those with Cohen’s d > 1 reported in other reviews (summarized in Thibaut et al., 2015).
This is an appropriate foundation for the comparisons of methods that motivated the study, though it was expected given the ratios reported for the overlap in par-ticipants with Carolus et al. (2014). In the latter study, in which M50 was deter-mined by dipole fitting by trained raters, ratios were 0.39 for HC, 0.50 for CHR, and 0.61 for FA. No HC vs. CHR differences were observed for any M100 or single-stimulus measure, evidence that the traditional P50 or M50 gating abnor-mality is not driven primarily by N100 or M100 or by only S1 or only S2 responses.
The second hypothesis, that quantification methods differ in the magnitude of effect sizes, was not confirmed. The 12 M50 gating effect sizes in Table 2.2 comparing HC to one of the SZ groups ranged from .53 to .71.(The respective effect sizes for the overlapping sample reported in Carolus et al., 2014, were .79
to .87). The 12 M50 gating quantifications were not markedly discrepant in effect size, and the Friedman test of heterogeneity did not find significant differences.
The magnitude of effect sizes the HC vs. SZ group difference was largely con-sistent across quantification methods: for the same 12 M50 gating effect sizes in Table 2.2, HC and SZ were consistently distinguishable in M50 gating, with sim-ilar medium-to-large effect sizes across quantification methods. No HC vs. SZ differences were found for M100 or for S1 or S2 responses analyzed individually.
Thus, regarding the primary motivation for the study, M50 gating and the SZ gat-ing deficit were robust to differences in gatgat-ing quantification method.
Hypothesis 3, that CHR/FA differences, if any, differ by quantification method, received no support, in that all six effect sizes in the bottom row of Table 2.2 were small (.07 to .27) and nonsignificant. This result is consistent with the characterization of the P50/M50 sensory gating deficit as a core feature of schiz-ophrenia psychopathology evident early in the course of the disorder and not merely a consequence of chronic illness and/or multiple inpatient treatments.
Evaluation of multiple quantification methods in a given data set can com-pensate for some issues of data quality, as they have different strengths and weaknesses. For example, individual head shape and sensor-position differ-ences can be compensated for by using individualized peak sensor selection or by analyzing in source space, and the impact of differences in overall sensor sig-nal strength can be mitigated by using a ratio rather than a difference score or by using source analysis. Moreover, as the meta-analysis by de Wilde et al. (2007) emphasized, methodological differences between studies, which may contribute to inconsistent results, make it difficult to evaluate the stability of sensory gating effects across studies, such as to establish it as an endophenotype for schizo-phrenia. The present independence of effect sizes and statistical results from the
quantification methods evaluated here encourages the study of sensory gating deficits as a core feature of schizophrenia psychopathology and in particular as an endophenotype (Miller & Rockstroh, 2013; Thibaut et al., 2015).
Beyond the general consistency of the present MEG results with the gating literature, which is largely EEG-based, the present source analysis replicated the left-hemisphere dominance of the HC vs. CHR contrast reported in previous MEG gating studies. Specifically, only left-, not right-hemisphere quantifications showed a HC vs. CHR difference (left vs. right comparison not significant for M50 minus baseline ratio, t(120)<1, p>0.1, nor for M50 minus M100 ratio, t(120)<1, p>0.1, but highly significant for S1-S2 M50 Difference, t(120)=5.62, p<0.001).
Similar to the source analysis, sensor-space results confirmed stronger HC vs.
CHR contrasts for left-hemisphere sensors.
Several limitations of the present study can be noted. There are significant advantages to the present comparison of gating quantification methods from a single large data set, thus avoiding cross-study differences in participants, re-cording, and preprocessing as a contributor to inconsistent results, but the gen-eralizability of any one data set are necessarily limited. Systematic simulations to evaluate the effects of preprocessing options and their interactions with cation method remain to be done. Still, present results suggest that that quantifi-cation method alone is not necessarily a barrier to successful use of P50/M50 in studies of normal function or psychopathology.
A second limitation is that the present comparisons were restricted to methods of quantification of sensory gating and did not directly address differ-ences in how EEG and MEG signals appear at the scalp due to their somewhat different biophysical properties, which could be a source of inconsistent results.
These differences were addressed in several indirect ways. First, for some
analyses, MEG sensor data were averaged across hemispheres to approximate midline EEG recordings. Second, we adjusted quantification methods: unlike in EEG recordings, in MEG recordings vertically oriented lateralized sources (such as Heschl’s gyrus for M50; though see Edgar et al., 2003, for evidence of cross-subject variability in orientation) typically produce magnetic fields, and thus MEG scalp components, with opposite polarities over homologous sites. (Conversely, EEG would do this with sources oriented left to right, and in such a case lateralized MEG recording would not be subject to such a polarity reversal.) This difference from typical EEG P50 studies was addressed here by identifying M50 and M100 as components having complementary polarities. Third, the possibly less consistent placement of the MEG helmet compared to a fixed EEG net was counteracted by choosing individual peak sensors or by source reconstruction.
These strategies notwithstanding, EEG and MEG provide both redundant and complementary information, so a direct comparison of results would be limited by inherent differences, e.g. better sensitivity to radial (EEG) vs. tangential (MEG) sources. MEG provides advantages, such as better source reconstruction (Edgar et al., 2003), better test-retest reliability (Lu et al., 2007), and better differentiation of hemispheres, whereas EEG is cheaper to purchase and use and much more widely available. Even though the neural activity of interest in paired-click gating studies manifests in both electrical and magnetic fields, so that EEG and MEG measurements are equally of interest, most of the literature has used EEG. Thus, the present study would be fruitfully complemented by a similar EEG study.
Another possible source of differences across studies is stimulus proper-ties. Characterization of the intensity of such a brief auditory stimulus in a way that would be replicable cross-lab is quite difficult. Stimulus rise times are often not reported and presumably not measured or controlled (they were not in the
present data set). Even measuring dB with such a brief stimulus is challenging.
In any case, the generalizability of present results to stimulus properties other than those used in the present data set is unknown. However, the similarity of present sensor-space gating ratios to those in the literature provides some evi-dence that results apply to other contexts.
Age, medication, nicotine, and education may be considered potential con-founds, both within a study that compares groups and across studies, especially because FA patients had been receiving neuroleptic medication for a shorter pe-riod than had CHR patients. However, taking these variables into account as co-variates did not change the pattern of results (see Supplementary Table 2.1).
Two final limitations concern the time course of schizophrenia. The robust-ness of sensory gating deficits over the course of a disorder cannot be adequately evaluated by cross-sectional comparison of samples presumably at different stages of the disorder, with markedly different treatment histories. In addition, the inclusion of FA patients does preclude these patients having experienced psy-chotic symptoms before admission. That is, FA participants may not represent first-episode phenomena. Nevertheless, consistency of findings across two sam-ples that differ in the amount and duration of treatment suggests that the gating deficit is not a result of long-term medication or other aspects of chronicity. Thus, present results are in line with evidence of the sensory gating deficit in first-epi-sode patients (Yee et al., 2010) and support its interpretation as an index of treat-ment-independent psychopathology.
The comparison of several measures within the same data set suggests independence of auditory gating effects from the specific method of M50 quanti-fication, supporting sensory gating as a robust assay of psychological and neural function. Moreover, the comparison of these measures in patient groups differing
in stage of illness supports treating abnormal gating as an index of fundamental psychopathology in schizophrenia and candidacy as an endophenotype. Alt-hough absence of statistical differences among quantification methods cannot constitute proof, present results, based on larger sample Ns than many past P50 and M50 studies, indicate that differences in quantification methods do not con-tribute critically to sensory gating differences reported in the literature.